An agent-based modeling of rescue operations for the evaluation of short-range flash flood forecasts

This paper presents the application of an agent based model to evaluate how short-range flood forecasts may improve the emergency management of a severe flash-flood event. The considered flood hit several inhabited areas in the Aude River basin, south-eastern France, in October 2018. Particularly, the capacity to trigger timely rescue operations at the right places is evaluated and compared, depending on the input information used in the emergency decision process: from rainfall observations up to flash flood impacts forecasts at the river reach scale. To achieve this evaluation, the field operations of firemen rescue teams are simulated in an agent-based model, which provides a detailed description of the complexity of the emergency situation: location and timing of flood damages, limited number of rescue teams, traveling times, decisions taken under uncertainty about the future evolution of the event. The flash flood impacts forecasts involved in the decisions are obtained using (i) three different short range (0-6h) quantitative precipitation forecast (QPF) products or a naive zero future rainfall scenario, (ii) a distributed hydrological model and (iii) a simple impacts model evaluating the number of flooded buildings. The presented results confirm that the efficiency of the rescue operations is generally improved when using QPFs as input of the decision chain, due to the increased anticipation. But they also illustrate how this added value can be highly altered by the combined effects of forecast uncertainties (false alarms) and limited available rescue means. In a fictive situation where rescue means are extremely limited, using a zero future rainfall scenario proves to be almost as efficient as using QPFs. • Short range flood forecasts are evaluated for the Aude 2018 flash-flood. • An Agent Based Model is implemented to represent rescue operations during the flood. • The rescue operations can be triggered based either on forecasts, or observations. • Forecast-based rescue operations are more efficient despite forecast uncertainties. • The value of forecasts decreases when operational rescue means are limited.